Locked out of ChatGPT? Your EU rights, and how to back up your chats now

What's going on

Since November 7, ChatGPT users have been reporting that their accounts became unusable overnight. Logging in only shows "account_deactivated", and most got no email warning first. According to dev.ua, Kod.ru and BigGo Finance, $200-a-month Pro subscriptions are affected too, along with shared Team accounts and free API accounts, some of them active for five years. Losing access means losing chat histories, projects, the memory feature (the notes ChatGPT keeps about you) and, for some people, years of journal entries and work notes. As of November 8, OpenAI hadn't published a statement or any figures on how many users are affected.

Our take: if you keep your work in a chatbot, you don't own it, you're renting it. EU law gives you some leverage, but it works slowly and is in places weaker than often claimed. Backing up, on the other hand, takes just a few minutes.

What we know about the ban wave

Only individual cases are documented; there's no overall picture. One Pro customer from OpenAI's program for security researchers had protected his account with a physical security key. He suspects his VPN connections through various countries triggered the ban, but stresses himself that this is only a guess. Support gave him no reason and pointed him to the appeal process. According to Kod.ru, a Reddit user was banned over an alleged link to a previously banned account, even though he says he never had a second account. His appeal was rejected after about an hour. Another user who has been paying for four years got his first rejection within minutes and the second after about a minute. According to dev.ua, OpenAI has also started limiting how many paid accounts are allowed per device and per bank card.

On the Chinese forum V2EX, users suspect the main targets are reverse proxies: intermediary servers through which many people use a single account at once, often resold. There's also talk of guilt by association there: one user reportedly lost more than ten Pro accounts in one go. At the same time, many forum members say their accounts are working fine. V2EX is also where speculation surfaced that Cat Wu had something to do with it. The former head of product for Claude Code moved from Anthropic to OpenAI, where, according to The Next Web, she leads product management for the core products. There's no evidence of any connection to the bans.

This isn't the first wave. Back in June and July, the OpenAI Community forum filled up with similar cases. One user was told his appeal had been filed "more than 180 days" after the ban, even though the ban was only four days old. Other accounts were restored after a successful appeal, but only as free accounts. Several bans cited distillation as the reason, meaning harvesting model outputs to train your own model. In a GitHub issue, a Pro customer describes how a bot rejected his appeal within seconds. Only escalating through the help chat got his account back.

What the DSA actually gives you

On August 31, 2026, the European Commission designated ChatGPT a Very Large Online Search Engine (VLOSE) under the Digital Services Act. It calls ChatGPT a "hybrid service" because it searches the web. By OpenAI's own figures, an average of 159.1 million people per month use ChatGPT search in the EU. By January 2027, OpenAI has to meet the additional obligations: risk assessments, annual independent audits, more transparency, terms of service in all official EU languages, and a plain-language summary that also lists the ways to file complaints (Art. 14(5)).

The rights that banned users are hoping for, however, depend on the type of service, not its size:

Search engines aren't explicitly mentioned in Art. 17, 20 and 21. The Commission designated ChatGPT as a search engine, not a platform. An arXiv paper counters that ChatGPT is also a hosting service because it stores prompts and custom GPTs. That hasn't been settled.

Our take: there's a good case for a duty to give reasons under Art. 17, but it would then have applied since 2024, not just from January 2027. When it comes to the complaint system and dispute settlement (Art. 20/21), banned users are on shaky legal ground. And with Custom GPTs ending on December 11, the GPT Store goes away too, the part of ChatGPT that most resembled a platform. What definitely applies is the duty of care under Art. 14 and the right to complain to the regulator under Art. 53.

Data protection: your data is still your data

The GDPR is stronger here, because it applies whether or not you can still log in.

• Right of access (Art. 15): You're entitled to a copy of all the data OpenAI processes about you. The response is due within one month; in exceptional cases the deadline can be extended by two months. According to OpenAI's privacy policy, the company may retain data from banned accounts. So your data most likely still exists.
• Data portability (Art. 20): What you provided yourself, meaning prompts and uploads, has to be handed over in a machine-readable format. Whether the AI's responses count as well is disputed.
• Automated decisions (Art. 22): A ban with significant consequences made solely by an algorithm is only allowed in exceptional cases, for example when it's necessary to perform a contract. Even then, you're entitled to human intervention, to put forward your point of view and to contest the decision. The Uber case shows how strictly this can be interpreted: in 2023, the Amsterdam Court of Appeal ruled that automated deactivations of driver accounts were solely automated because human oversight was "little more than symbolic". And under a February 2025 CJEU ruling (C-203/22), how an automated decision was reached has to be explained in an understandable way.

A forum case from June shows how this plays out in practice: OpenAI's privacy portal just reported "account deleted or deactivated", and support sent a generic article on data export. Back in 2024, the privacy group noyb complained to the Austrian data protection authority that OpenAI's responses to access requests were incomplete. You can file a complaint with your local data protection authority (in Germany, the one for your state). Because OpenAI Ireland is the contracting party for the EU, Ireland's Data Protection Commission (DPC) takes the lead. That takes time.

German contract law applies on top of that. In the 2021 Facebook case, Germany's Federal Court of Justice ruled that users must be informed and given a chance to respond before an account suspension, followed by a new decision (III ZR 179/20). No court has yet ruled on whether that carries over to ChatGPT.

How to back up your data before it happens

The ChatGPT and Claude exports only work while you're logged in. According to Anthropic's help pages, you can't trigger a new export once an account has been closed. With Google, some suspensions still allow a data download, but not for serious violations. And keep in mind: a Google ban hits your entire account, Gmail included.

Here's how to get a backup in five minutes:

1. Start the export, download the archive before the link expires, and store it outside the provider's cloud.
2. Copy your memory entries as text, either through the settings or by asking the bot to output them verbatim. Claude recommends exactly that.
3. Back up the instructions and knowledge files of your custom GPTs and Gems separately. For GPTs you'll need to do this before December 11 anyway. If you convert them into plugins, you can download them as a ZIP.
4. Repeat this regularly, say monthly, and save important results straight away in common formats like Markdown or PDF.

If it happens to you

Take screenshots of the error message and note the time and, if available, the request ID. File an appeal via the link in the ban email, the form or the help chat. In it, explicitly request human review and a specific reason. At the same time, submit a request for access and a copy of your data under Art. 15 and 20 GDPR, and ask them to preserve your data. If you pay through Apple or Google, cancel the subscription yourself: according to OpenAI's help pages, a ban only stops auto-renewal for subscriptions purchased directly from OpenAI. Don't create a new account, because a link to a banned account is itself grounds for a ban. If you haven't heard back after a month, contact your data protection authority and your national Digital Services Coordinator (in Germany, the Bundesnetzagentur).

The bigger picture

The only public figures show how lopsided the balance of power is: Anthropic says it banned 11.4 million accounts in the first half of 2026. It received 398,000 appeals, about 3.5 percent, and 42,000 of them succeeded. That's about 10.6 percent of appeals and roughly 0.4 percent of all bans. We couldn't find comparable figures from OpenAI. Anthropic, too, banned users in Hong Kong and Russia who accessed its services via VPN in October. So the whole industry relies on automated detection, and the mistakes hit paying customers.

Our take: for people in the EU, the GDPR is the most reliable lever, especially the right of access and the right to human review. From January 2027, the DSA mainly brings audits and more transparency. It will only deliver clear complaint and dispute settlement rights for ChatGPT users once it's settled whether ChatGPT is also a hosting service or a platform. Until then, one simple rule applies: anything that exists only in your chat history can vanish overnight. An export at the start of every month is the cheapest insurance you can get.